Autonomous Car Progress

[diplomat33]

How many rockets have you sent to the ISS by the way?

Sending rockets to the ISS is a completely different problem than L5 autonomy. Sending rockets into orbit is a solved problem that we've been doing since the 50's. Nobody has solved L5 autonomy yet.

 

[Cheburashka]

Sending rockets to the ISS is a completely different problem than L5 autonomy. Sending rockets into orbit is a solved problem that we've been doing since the 50's. Nobody has solved L5 autonomy yet.

Not true. Musk also invented the subway and tunnels, so he clearly know everything there is about FSD and Tesla is way ahead. How many tunnels have you built?

 

[DanCar]

... I am telling you Tesla are further ahead than you guys think behind closed doors. ...

That they has been bounced around for 4 years and will be around for 5+ more years.

 

[daniel]

Bloomberg - Are you a robot?

I am telling you Tesla are further ahead than you guys think behind closed doors.

I also don't think Lidar is necessary for L5.

Some of you are going to say Elon has been wrong before and it's true but nobody's perfect.

How many rockets have you sent to the ISS by the way?

Elon is an optimist raised to the power of ridiculous. Were it not for this we would not have these cars. I'd never have driven my Roadster as my daily driver for seven years. I'd never have had my Model 3 whose (misnamed) Enhanced Autopilot makes it the safest car I've ever driven. SpaceX would not be sending astronauts to the ISS, the first private company ever to send humans into space. Were it not for Powerwall I would not have been able to put solar panels on my roof because my utility is fully-subscribed for net-metering and won't buy my excess daytime power. I admire him and his drive and dedication tremendously. It needed a ridiculous level of optimism to make all these things happen because a less optimistic person never would have dedicated so much time and money to it. A less optimistic person would have stuck with tried-and-true business models.

But this same optimism makes him think that L5 autonomous driving is a much easier problem to solve than it really is. I admire Musk and I love Tesla and I think they will achieve fully autonomous driving. I just don't think that it will happen as soon as Musk thinks. And I don't think it will happen on the hardware in Tesla's 2020 cars. There remains a long road ahead for fully autonomous driving, and Tesla will get there in part because Musk thinks (mistakenly) that it's just around the corner. We need such optimists. But we let ourselves in for disappointment if we believe Musk's timeline predictions.

Level 5 is just perfect Level 2.

Not true. There are fundamental differences. At Level 2, by definition, the car need not be capable of determining when it can and cannot handle a situation. At Level 2 the car is allowed to drive itself into situations it is not capable of recovering from, because the driver is responsible. Also, Level 2 need not be capable of more than speed control and lane-holding. It can have more features, but to qualify for Level 2 doesn't need them. Level 3 also doesn't need more features than speed control and lane keeping, but must be able to determine with a few moments of advance notice, when it cannot handle an upcoming situation. That is a HUGE step! Level 4 is an even BIGGER step because it does not require a human in the driver's seat. And Level 5 is a whole 'nuther beast entirely because it doesn't even need driver controls.

You ain't getting from Level 2 to Level 5 in half a year. From here to a car that does not need to have driver controls is at least a decade.

And of course, nothing is "perfect." Level 5 will not be perfect. There will be deaths in Level 5 cars. As long as there are significantly fewer than the present thirty-thousand a year in the U.S. alone it will be good enough.

 

[mikes_fsd]

You ain't getting from Level 2 to Level 5 in half a year. From here to a car that does not need to have driver controls is at least a decade.

There's this problem of regulations that need to catch up.
So, the user controls (i.e. steering wheel, pedals) will be there and the requirement of a user to be in the driver seat until those regulations are re-written/fixed.

What the FSD feature set will give Tesla is the ability to validate with a "free safety driver" (i.e. you the owner) their NN's in real world situations and find those weird scenarios for the long tail that Musk brought up in the interview.

 

[Soda Popinski]

Sending rockets to the ISS is a completely different problem than L5 autonomy. Sending rockets into orbit is a solved problem that we've been doing since the 50's. Nobody has solved L5 autonomy yet.

Agreed as a self-acknowledged video game space expert. It was a lot harder writing my Munar rover's autopilot program (using rudimentary Lidar to avoid boulders) than to do a precision landing on top of a derelict space craft.

I have been saying and will say it agin. Here is guy who lands boosters on barges in the middle of the the ocean and send people to the ISS for pennies to the dollar for price it used to be. Talk about hard things.
Yes, he was of in his time but Elon alway comes through eventually. His timing has become better recently, especially short time prognoses.
Don't underestimate him.

I think people are arguing about two things. 1) Will Tesla achieve FSD with current hardware, and 2) Will they beat their competitors to FSD?

For number 1, my gut tells me 70-80% they will, but it won't be great FSD (like NoA functions, but I can't use it in Los Angeles traffic). I don't think they need Lidar, but I think they need slightly better camera coverage, as the B-pillar cameras aren't forward enough for cross traffic (which could be fixed with a 2nd camera in the repeater area or integrated into the headlight housing).

For number 2, I'd say they lost already for L4, and guessing probably lose in first to market wide-spread FSD (still either L4 in every major city or L5), but may scale up to 1 million car faster.

There's this problem of regulations that need to catch up.
So, the user controls (i.e. steering wheel, pedals) will be there and the requirement of a user to be in the driver seat until those regulations are re-written/fixed.

What the FSD feature set will give Tesla is the ability to validate with a "free safety driver" (i.e. you the owner) their NN's in real world situations and find those weird scenarios for the long tail that Musk brought up in the interview.

There's more to it than just regulation. We've got hints from Karpathy's talks on the great strides Tesla has made in perception, but we haven't seen squat on progress on driving policy. Turns out that's also hard to figure out.

 

[bd7349]

Here is a full quote from Elon from WAIC in more context:

“I think there are no fundamental challenges remaining for Level 5 autonomy. There are many small problems. And then there’s the challenge of solving all those small problems and then putting the whole system together and just keep addressing the long tail of problems. So you’ll find that you’re able to handle the vast majority of situations. But then there will be something very odd. You have to have a system figure out a train to deal with these odd situations. This is why you need real world situations. Nothing is more complex and weird than the real world,”
Tesla's Elon Musk talks Autopilot and Level 5 Autonomy at China AI conference

I am sorry to sound "anti-Tesla" but that first sentence sounds incredibly naive or arrogant to me. There are really no fundamental challenges remaining for L5? It's all just small problems from here on out? Who knew!

I’m not saying Elon is right here, but I’ve noticed that you seem to base a lot of your comments on what our current version of Autopilot can handle/understand. It’s been shown both in the talk Andrej gave in Feb 2020 and in his CVPR video that what they’re working on internally and what we have are two very different software branches. You can see this in how the birds-eye NN predictor sees the world and how much more accurate it is compared to how our current sees the world (falls apart when it tries to make sense of an entire scene like an intersection).

Not only that, but based on the CVPR video it shows their birds-eye view predictor handling multi-lane intersections and predicting their layouts incredibly well. This was not show in the Feb 2020 talk, so it seems they’ve made significant progress from Feb to June and it only stands to reason that they’re making good progress on other parts of FSD as well since they clearly aren’t standing still. Coupled with Elon’s recent comments about releasing the new Autopilot software stack in 2-4 months with “a lot of new functionality” and this just further reiterates that what we’re running is all based on an old stack that they are barely even working on anymore.

I know you’ve watched all these videos and understand that these things are being worked on, but it seems like you still think Tesla will continue to stick with their L2 Autopilot system. What makes you think that despite the progress they seem to be making?

 

[daniel]

There's this problem of regulations that need to catch up.
So, the user controls (i.e. steering wheel, pedals) will be there and the requirement of a user to be in the driver seat until those regulations are re-written/fixed.

What the FSD feature set will give Tesla is the ability to validate with a "free safety driver" (i.e. you the owner) their NN's in real world situations and find those weird scenarios for the long tail that Musk brought up in the interview.

Level 5 by definition needs no human in the car. It's not Level 5 until it can do that. And Level 4 needs no human in the driver's seat. Agreed that At levels 2 and 3 Tesla can gather real-world driving information with an unpaid driver (the owner) in the driver's seat.

But regulatory approval will not be an obstacle. Once they've demonstrated that Level 4 or 5 is safer than a human driver, they'll have the ultra-powerful insurance industry, with all its lobbying power, on their side. And it won't be Tesla against the world. It will be all the car companies, with the insurance industry, against a few neo-Luddites.

They will get there. Just not as soon as the most optimistic are expecting.

And I'll upgrade my car as soon as they do. Probably by buying the latest model, since I expect to be ready for a new car when the time comes. But if they "flip the switch" to level 5 next week, I'll go on line and pay for the FSD package immediately.

 

[Soda Popinski]

Not only that, but based on the CVPR video it shows their birds-eye view predictor handling multi-lane intersections and predicting their layouts incredibly well. This was not show in the Feb 2020 talk, so it seems they’ve made significant progress from Feb to June and it only stands to reason that they’re making good progress on other parts of FSD as well since they clearly aren’t standing still. Coupled with Elon’s recent comments about releasing the new Autopilot software stack in 2-4 months with “a lot of new functionality” and this just further reiterates that what we’re running is all based on an old stack that they are barely even working on anymore.

Keep in mind, the bird's eye view from integrating all the camera data along with the per-pixel distance data from Pseudo-Lidar basically puts Tesla where Waymo was several years ago. They may have "solved" perception for the most part, but so have others, and they're still both dealing with the long tail of weird stuff to perceive, and the whole kit and kaboodle of driving policy / decisions that we haven't seen much about besides Autopilot.

 

[bd7349]

Keep in mind, the bird's eye view from integrating all the camera data along with the per-pixel distance data from Pseudo-Lidar basically puts Tesla where Waymo was several years ago. They may have "solved" perception for the most part, but so have others, and they're still both dealing with the long tail of weird stuff to perceive, and the whole kit and kaboodle of driving policy / decisions that we haven't seen much about besides Autopilot.

True, but what I’m saying is since they’ve shown they’re not standing still on solving perception, that gives me reason to believe they’re also not standing still regarding the other parts of self driving (driving policy, planning, etc).

My theory is I think they’re leading up to a big reveal of a brand new Autopilot in the next 2-4 months. I don’t think they’ve showed off those other aspects of self driving like Waymo has because then they’d have to reveal how their in development system is working and that would ruin the big reveal that’s in a few months.

To be clear I don’t think that what they reveal will be L5, that would be truly astonishing if it were the case. But I do think we’ll see a completely new Autopilot that will be much more hands off... L4 with supervision at first seems likely, but even that would be a massive leap over what we have now.

 

[Soda Popinski]

True, but what I’m saying is since they’ve shown they’re not standing still on solving perception, that gives me reason to believe they’re also not standing still regarding the other parts of self driving (driving policy, planning, etc).

That's a good point. My concern is, we've seen plenty of evidence of their progress on perception, but little on driving policy. Perhaps we're biased because of the nature of the conferences Karpathy was speaking at. Or maybe they need to "solve" perception better before they can start on developing better driving policy.

My theory is I think they’re leading up to a big reveal of a brand new Autopilot in the next 2-4 months. I don’t think they’ve showed off those other aspects of self driving like Waymo has because then they’d have to reveal how their in development system is working and that would ruin the big reveal that’s in a few months.

An interesting point I hadn't considered that Tesla has compared to Waymo is, Tesla can start rolling out "FSD" features to "feature complete" and still give a lot of value to drivers without needing to be at 99.99999% since every car will have a "safety driver." Waymo really needs to hit all those 9s because the passenger is the customer, and not the driver, and they don't have a business model until they ditch the safety driver.

 

[diplomat33]

I’m not saying Elon is right here, but I’ve noticed that you seem to base a lot of your comments on what our current version of Autopilot can handle/understand. It’s been shown both in the talk Andrej gave in Feb 2020 and in his CVPR video that what they’re working on internally and what we have are two very different software branches. You can see this in how the birds-eye NN predictor sees the world and how much more accurate it is compared to how our current sees the world (falls apart when it tries to make sense of an entire scene like an intersection).

Not only that, but based on the CVPR video it shows their birds-eye view predictor handling multi-lane intersections and predicting their layouts incredibly well. This was not show in the Feb 2020 talk, so it seems they’ve made significant progress from Feb to June and it only stands to reason that they’re making good progress on other parts of FSD as well since they clearly aren’t standing still. Coupled with Elon’s recent comments about releasing the new Autopilot software stack in 2-4 months with “a lot of new functionality” and this just further reiterates that what we’re running is all based on an old stack that they are barely even working on anymore.

I know you’ve watched all these videos and understand that these things are being worked on, but it seems like you still think Tesla will continue to stick with their L2 Autopilot system. What makes you think that despite the progress they seem to be making?

Yes, I've seen the CVPR videos (I started that thread). I am aware of the NN work that Tesla is doing. And the work looks really good actually. But it's all perception stuff. Tesla does not seem to be doing a lot of work with planning and driving policy which is a huge part of autonomous driving.

For example, the NN prediction of intersections just deals with the problem of "seeing" an intersection, it does not really deal with the problem of handling the traffic around the car. For example, can the car safely handle an unprotected left turn where a cyclist cuts in front while you are half way into the intersection? Can the car handle an intersection that is under construction where the car needs to temporarily move over into the incoming traffic lane but there is a large truck partially blocking the view? These are the kind of driving scenarios that require more than just seeing an intersection, the car also needs that planning and driving policy to figure out the safest to way to handle what other vehicles are doing.

So again, it's awesome to see the cool perception stuff that Tesla is doing but we have no way of knowing how good the features will actually work in reality, especially in scenarios like the ones I mentioned. I don't think we can make "forward looking statements" and say that just because Tesla has NN that can predict intersections like 99% of the time, that out cars will be L5.

No, I do not think that Tesla will stay at L2 forever. I have no doubt that the rewrite and these new NN that we got a glimpse of at CVPR will add new functionality. Absolutely! In fact, I think we will get "turning at intersections" at some point. Eventually, Tesla will get to L3+. But getting to L5 is a huge task because of all the driving cases that need to be solved.

The reason I bring up current AP is because that is something I can actually judge. It does not require guesswork or forward looking statements. I can test how AP handles specific driving scenarios. And certainly, when Tesla releases new features I will eagerly test those. and if AP is able to handle an unprotected left turn with a cyclist that cuts in at the last second without driver intervention, I will be the first to applaud Tesla. I want to judge Tesla based on what they actually deliver to my car, rather than NN work that may or may not translate into a good feature.

Bottom line: L5 means being able to handle all roads, all driving cases, day and night, with no human intervention. That's a tall order. Tesla is not there yet. I am not saying Tesla won't make progress or even that Tesla won't achieve L4 or L5 someday. I am just saying let's not jump the gun. Being able to predict intersections is cool but there is a lot more that goes into L5.

 

[Dutchie]

Sending rockets to the ISS is a completely different problem than L5 autonomy. Sending rockets into orbit is a solved problem that we've been doing since the 50's. Nobody has solved L5 autonomy yet.

But landing boosters on barges in the middle of the ocean?

Come on, guys! There are hundreds of people at Tesla working on FSD. Most of the time of Elon at Tesla is dedicated to FSD. I have yet to hear anyone who actually is working or worked at Tesla doubt the progress and achieveability of the end goal. Most of the R&D budged is going to FSD.
Do you really think he is making this stuf up and he doesn't know what he is talking about? Do you really think he risks his company due to breach of contract for people who bought cars with FSD? They even increased the price of FSD recently. Do you guys really think they would have done that if they would not have show something soon?

 

[mikes_fsd]

So again, it's awesome to see the cool perception stuff that Tesla is doing but we have no way of knowing how good the features will actually work in reality, especially in scenarios like the ones I mentioned. I don't think we can make "forward looking statements"

Every video you post from Waymo, Cruise et al is a forward looking statement.

So, Tesla just needs to publish a new shiny demo video so you can approve of their driving policy?

Just like people did not know that every Falcon 1 rocket ( first 5 launches) had a parachute to bring the booster down for a soft water landing, we do not know how much effort and resources are being spent on the driving policy, but it sure as hell is not being "overlooked" because "Elon didn't provide a demo video".

 

[diplomat33]

Every video you post from Waymo, Cruise et al is a forward looking statement.

No, they are not. They are videos of what Waymo, Cruise et al have NOW.

 

[ZeApelido]

Not true. Musk also invented the subway and tunnels, so he clearly know everything there is about FSD and Tesla is way ahead. How many tunnels have you built?

You're gonna believe he did invent tunnels when he bores a straight path to Santa Cruz for you

 

[john5520]

But landing boosters on barges in the middle of the ocean?
Do you really think he is making this stuf up and he doesn't know what he is talking about? Do you really think he risks his company due to breach of contract for people who bought cars with FSD? They even increased the price of FSD recently. Do you guys really think they would have done that if they would not have show something soon?

Making unfulfilled promises is nothing new with Elon. According to past statements, we're supposed to be at L5 already. Sure, the re-write is looking promising and I can't wait to see what's around the corner, but at the same time, knowing what I know now about L5 and its requirements, I've tamed my expectations to match reality.

 

[mikes_fsd]

No, they are not. They are videos of what Waymo, Cruise et al have NOW.

That's not available to anyone. great!

 

[diplomat33]

That's not available to anyone. great!

Waymo is available to the public.

 

[Cheburashka]

You're gonna believe he did invent tunnels when he bores a straight path to Santa Cruz for you

He wouldn't be the first! There was already a railroad that used tunnels that went from LG to Santa Cruz in the 30s...